Fault Diagnosis Method for High-speed Train Gearbox Bearing Based on Improved VMD and Temperature-vibration Feature Fusion

Abstract

Objective Existing methods for high-speed train gearbox bearing monitoring and diagnosis in China often rely solely on temperature or vibration data. Rely solely on a single temperature data point may result in missing early fault information of key components, while only vibration data may struggle to support identification of faults under complex coupling conditions. Therefore, it is necessary to combine temperature and vibration data to develop a fault diagnosis method for gearbox bearings with temperature-vibration features. Method To determine the decomposition parameters of VMD (variational mode decomposition) method, a weighted kurtosis coefficient indicator is introduced. Combining LMD (local mean decomposition) and VMD methods, a new approach for processing raw vibration data and extracting fault features is proposed. Based on the improved VMD method, LLE (locally linear embedding) feature dimensionality reduction method, and BP (back-propagation) neural network, a method for temperature-vibration feature fusion in bearing fault diagnosis is proposed. Time-domain features and temperature features are used as inputs to establish the temperature-vibration feature fusion bearing fault diagnosis model. Using a high-speed train rolling bearing test bench, fault simulation tests are conducted on gearbox bearings of a certain type of high-speed EMU (electric multiple units) in China, and relevant vibration data are collected to validate the effectiveness and feasibility of proposed model. Result & Conclusion The proposed fault diagnosis method for gearbox bearings achieves an average identification accuracy of over 98% for normal state, outer ring fault, and rolling element fault.

Keywords

• high-speed train
• gearbox bearing
• vmd